As a non-contact temperature measurement tool, infrared thermal imagers offer significant advantages in hot stamping processes. However, their measurement accuracy is susceptible to multiple factors, including surface emissivity, observation angle, and target temperature. A temperature measurement optimization method based on dynamic emissivity compensation is proposed. Projection measurement technology is used to accurately obtain the spatial angular parameters of complex curved parts. Then, the effect of observation angle and temperature value on temperature measurement deviation is quantitatively analyzed through experiments. A machine learning algorithm is employed to construct a nonlinear mapping model between emissivity and multidimensional variables, enabling intelligent compensation of dynamic emissivity parameters. Experimental results show that after compensation, the temperature measurement system error can be stably controlled within the range of ±1.5 °C, improving accuracy by 60% compared to the fixed emissivity mode. This method provides an effective solution for the application of high-precision infrared temperature measurement in intelligent manufacturing scenarios.